tacit collusion 2012 - Universidad Católica del Norte · tacit collusion, under predictable demand...
Transcript of tacit collusion 2012 - Universidad Católica del Norte · tacit collusion, under predictable demand...
Tacit Collusion in Housing Markets
The Case of Santiago, Chile
Fernando Lefort∗and Miguel Vargas†
Facultad de Economıa y Empresa
Universidad Diego Portales
Abstract
In this paper we investigate the potential existence of tacit collusion in housing
markets using a detailed micro data base from Santiago, Chile. In order to perform
the test, we first split Santiago’s housing market into four different sub-markets
using hedonic price analysis and households socioeconomics characteristics. This
procedure is important because facilitates a more precise characterization of markets
and calculation of markups. Secondly, using a GMM panel data regression model
we run regressions, for each sub-market, correlating industry’s markups with the
aggregate level of activity. The main finding is that low and middle income sub-
markets present higher average markups and a pro-cyclical behavior. This finding
is consistent with a market where participants do not face capacity constraints and
behave strategically to sustain tacit collusion during increasing demand periods.
1 Introduction
It has been shown, (see for instance, Straszheim (1975)) that there is a strong relationship
between home ownership, access to credit, productivity in family- owned businesses, labor
∗[email protected]†[email protected]
1
market insertion and household income. Hence, a well-functioning, competitive housing
market can promote development, poverty reduction and improvements in quality of life
for citizens.
However, some particularities of housing markets may preclude the perfectly compet-
itive behavior of their participants. On the one hand, the special features of dwelling
units, particularly regarding location and quality heterogeneity, facilitate the existence of
monopolistic competition in housing markets, even in the presence of many competing
players.1
On the other hand, according to industrial organization theory, even a large number of
suppliers cannot guarantee competitive behavior of players due to the potential emergence
of tacit collusion, especially when, as it is the case in housing markets, multi-market
contact increases the frequency of the interaction between firms. 2
In general, the participants of a market operating either under monopolistic competi-
tion or tacit collusion will earn abnormal returns during some periods of time. Ching and
Fu (2003) empirically test this hypothesis for the Hong Kong urban land market and find
evidence of positive expected abnormal returns earned by developers. Although, monop-
olistic competition and tacit collusion are both departures from perfect competition, they
produce different private and social outcomes in the market. Furthermore, in theory, in
the absence of entry barriers only a collusive behavior will be able to maintain abnormal
returns in the long run.
Given the doubts cast by the economic literature regarding the level of competition
in housing markets, in this paper we investigate the potential existence of tacit collusion
1See for instance Taltavull de la Paz (2001).2As clearly stated by Ivaldi et al. (2003) collusion arises from dynamic interaction, a pervasive situation
in housing markets. For example, Straszheim (1975) indicates that variation in housing characteristics
and prices by location is a fundamental characteristic of the urban housing market, while Goodman
and Thibodeau (2003) point out that metropolitan housing markets are, in fact, segmented into smaller
submarkets, due to sector specific supply and demand factors. Furthermore, Bernheim and Whinston
(1990) show, using supergame analysis, that markets arranged in multi-markets, such as the housing
submarkets, facilitate collusive behavior in a wide range of circumstances.
2
in the Santiago de Chile housing markets. However, the economic empirical literature
has shown that, under tacit collusive behavior, i.e. in the absence of a smoking gun, it is
difficult to prove the existence of collusive behavior in a market.3
An important amount of empirical studies about tacit collusion have analyzed the
time pattern of mark-ups. The reason for this is that tacit collusion equilibrium may be
unstable and, hence, it should be expected to observe periods of high mark-ups followed
by periods with no abnormal returns. As an example, consider the situation where a
reduction in profits caused by an exogenous factor or the defection of some of the colluding
firms, triggers retaliation behavior by firms causing a period of low profits in the market.
Examples like the above have motivated researchers to empirically analyze the rela-
tionship between mark-up and the business cycle. This is because the pattern of mark-ups
during the business cycle may provide evidence of strategic behavior by firms. An addi-
tional difficulty faced by this line of research is that theory provides different conclusions
regarding how mark-ups should behave over the business cycle depending upon the as-
sumptions of the model.
Theoretical models of collusive behavior are framed under repeated games where firms
try to sustain collaborative high levels of profits through the threat of punishing defectors
increasing supply. In such a context, there are two key assumptions that shape the
theoretical relationship between mark-ups and the level of economic activity.
On the one hand, the predictability of demand conditions affects the ability to sustain
collusion. Intuitively, tacit collusion will be easier to sustain in booming markets, when
future profits are expected to be high and, hence, the expected cost of retaliation is also
high. Conversely, collusion is more difficult to sustain in declining markets because the
short run gain of defecting will tend to compensate the limited expected cost of future
retaliation.
The strategic behavior simply outlined above has two empirical implications. First,
in general, tacit collusion will be less sustainable in markets that are subject to demand
3See for instance Rapson (2009).
3
fluctuations especially when they are predictable.4 The second testable implication is that
tacit collusion, under predictable demand shocks and relatively homogeneous players will
tend to generate a pro-cyclical pattern of mark-ups.
The second key issue that may have an effect on the relationship between mark-ups
and the level of activity is the industry cost structure. In general, an asymmetric cost
structure will hinder collusion and condition market shares of participants.5 In theory,
however, capacity constraints have an ambiguous effect on the sustainability of collusion.
The reason is that although a capacity-constrained firm has less to gain when defecting a
tacit collaboration, it also has less retaliatory power against other defecting companies.
Because capacity constraints affect the strategic behavior of colluded firms, they may
also affect the relationship between mark-ups and the level of activity. Specifically, Fabra
(2006) shows that, if capacity constraints are sufficiently high, firms will find more difficult
to collude when facing increasing demand. Intuitively, when capacity constraints are
severe enough, the lack of excess capacity during a boom implies that the future costs
of being punished are low. Thus, the losses from cheating decrease even if collusive
profits are rising. In contrast, the emergence of excess capacity during a recession makes
the punishment threat more severe, and thereby induces an increase in the losses from
cheating even if collusive profits decline. Hence, a housing market where participant
companies have severe capacity constraints will tend to show counter cyclical mark-ups,
because firms will be able to coordinate better in times of decreasing demand.
In accordance with the above discussion, in this paper, we implement a test of tacit
collusion using a detailed housing sales’ data base from Santiago, Chile. The data includes
information about dwellings’ price, surface, number of bedrooms and bathrooms, and
location. We also have information on each location specific socioeconomic characteristics
and the quality of facilities available to households.
4This general idea was stated by Rotemberg and Saloner (1986) and Haltiwanger and Harrington
(1991). See also Ivaldi et al. (2003) for a clear discussion of this issue.5See Ivaldi et al. (2003) for a discussion in this issue.
4
In order to perform the test, we first split Santiago’s housing market into four different
sub-markets using hedonic price analysis and households socioeconomics characteristics.
This procedure is important because facilitates a more precise characterization of markets
and calculation of mark-ups. Furthermore, because there is more homogeneity among
dwelling units belonging to the same specific sub-market, sub-market markups are less
likely to reflect non-cooperative monopolistic competition. We find clear evidence of
positive average mark-ups in most sub-markets.
We, then, implement a test of the correlation between mark-ups and the level of activ-
ity for each sub-market, using a GMM panel data regression model regressing industry’s
sub-market markups against business cycle. The main finding is that low and middle
income sub-markets present both higher average markups and positive correlation with
the level of economic activity. This finding is consistent with a market where participants
do not face capacity constraints and behave strategically to sustain tacit collusion during
increasing demand periods.
Section 2 of the paper presents the methodology used to segment markets and calculate
mark-ups. Section 3 describes the Santiago housing market and the data used for the
econometric analysis. In section 4, we perform hedonic price regression in order to properly
identify the specific sub-market characteristics. In section 5, we use a GMM estimator
to obtain estimates of the correlation between mark-ups and economic activity for each
sub-market and analyze the overall results. Section 6 concludes.
2 Methodology and Data
The methodology proposed has been developed in order to test the presence of tacit collu-
sion in Santiago of Chile housing market. This test is based upon the works of Rotemberg
and Saloner (1986) and Green and Porter (1984), which establish that the relationship
between firms profits and business cycle will provide information about markets level of
competition. However, in order to do a more accurate analysis of the firms behavior we
need to identify sub-markets, given the particular features that these markets present.
5
After the sub-markets have been identified, the tacit collusion test will be performed for
each sub-market.
A simple algorithm of the methodology proposed here establishes the following steps:
• The estimation of a hedonic model for the city as a whole as a way to identify the
variables that are behind housing prices
• To cluster basic geographical units of analysis, like census tracks, according to a
criterion based upon households socio-economic characteristics. For instance to
cluster census tracks that have a similar average household income.
• Once the potential sub-markets have been defined the next step will be run a hedonic
regression for each one of them and then to test if the parameters estimated are
different between sub-markets.
• Once the sub-markets have been defined the firms markups will be estimated for
each sub-markets
• Finally, every sub-markets firms markups will be compare with the business cycle
in order to undertake the tacit collusion test
All these issues are discussed with further details in the following subsections.
2.1 Housing demand and hedonic prices estimation
Because dwellings and housing services are highly heterogeneous it is a difficult task to
estimate a generically demand function for them. Instead, dwellings can be decomposed
into its constituent characteristics and then estimates prices and elasticities for each one of
them. The way of doing that is using the hedonic regression due to Rosen (1974), which
faces the fact that observed choices over housing reveals to the researcher information
about the underlying preferences for these amenities or other characteristics of interest
(Taylor, 2008).
6
Let us consider that Pi is the dwelling price, which is a heterogeneous good, and xi is
a vector that includes its structural attributes of size and quality, characteristics of the
immediate neighborhood and indicators of its environment and accessibility. b is a vector
of parameters that must be estimated for the characteristics.
Pi = f(xi;b) + ui (1)
Having estimated the equation 1, it can be possible to predict the price of any dwelling
i whit attributes xi.
Pi = f(xi; b) (2)
For discrete characteristics the implicit price of the attribute kth can be calculated as
follows:
pk = f(xk + 1, x−k; b)− f(xk, x−k; b) (3)
and for the continuous case:
pk =∂f(xi; b)
∂xk
(4)
The implicit prices reveal the implicit marginal willingness to pay for an increment in
any of the dwellings attributes.
As Taylor (2008) points out, the hedonic price function has no theoretical guidance
for its specification, due to the fact that it is an envelope function. The most used
specification is a semi-log:
ln(Pi) = a+K∑
k=1
bkxki + ui (5)
The most common way of estimating 5 is by either OLS or maximum likelihood.
The set of the relevant attributes for price determination can be categorized into three
groups:
7
• characteristics of the dwelling and the lot
• features of the neighborhood, like crime rate
• the property locational characteristics, like the proximity to employment centers
2.2 Sub-markets definition
Despite the housing sub-markets, since their first appearance in the seminal work of
Maclennan (1977), have been widely studied in a theoretical framework, there is little
consensus about how sub-markets should be identified for applied studies (Alkay, 2008;
Royuela and Vargas, 2009).
In empirical works, sub-markets have been defined in different ways such as by de-
mand and supply factors, geographical characteristics, spatial characteristics, structural
characteristics and neighbourhood characteristics.
Researchers have offered different stratification schemes for their sub-markets defini-
tions: dwellings age, floor area, lot size, number of rooms, number of bathrooms, parking
lot, lift, wall material, roof material are given as examples of structural stratifiers. Also
socioeconomic characteristics and race have been used, and spatial dimensions as census
boundaries, neighborhood boundaries, municipal boundaries, school districts, inner and
outer urban areas. Income levels or household size in addition to neighborhood bound-
aries or inner and outer urban areas or construction type are examples of stratifiers of
joint influence.
Jones et al. (2004) defined sub-markets based on households intra-urban mobility. This
approach is an attempt to avoid researchers’ bias. In turn, within this structure different
approaches can be found too, such as travel-to-work areas and migration data.
Here the methodology introduced by Schnare and Struyk (1976), following the ex-
planation by Alkay (2008), is proposed. As sub-markets are not known in advance, the
first step must be to determine if segmentation exists. In order to do that, potential
sub-markets should be proposed, for instance, clustering census tracks with a similar av-
erage households income, and then to test if the parameters estimated for these potential
8
sub-markets are different between them. Second, if a segmentation structure is observed,
it must be tested if the resulting variation in prices is significant.
This test procedure can be split into three stages:
• First, to estimate a hedonic housing price function for each potential sub-market
in order to compare these potential sub-markets prices. If there are large and
significant differences in the estimated parameters of those potential sub-markets,
the differences might be accepted as evidence of market segmentation.
• Second, to compute the tests to establish whether significant differences exist be-
tween the sub-markets specific prices.
• Third, since the primary interest is in the price of housing instead in the price of
the individual housing characteristic, the difference between the whole market model
and sub-market models must be emphasized. Testing for the relative importance of
this variation the standard errors of the whole market model and the sub-markets
models can be compared.
2.3 Firms markups and the sub-markets level of competition
Machin and Van Reenen (1993) propose a procedure based upon super-games models
developed by Rotemberg and Saloner (1986) and Green and Porter (1984) to test the
extent of competition of an industry. These models have clearcuts predictions regarding
the behavior of markups over the business cycle: Rotemberg and Saloner (1986) predict
that markups should exhibit countercyclical behavior meanwhile Green and Porter (1984)
suggest pro-cyclical markups. The former prediction rely on the assumption that firms can
discriminate amongst aggregate and idiosyncratic shocks, whilst the latter prediction is
based upon the assumption that firms cannot do it. Therefore, if a systematic relationship
between profits and business cycle is found, it will be evidence of oligopolistic behavior.
The model estimated is the following:
9
yi,t = β0 + β1yi,t−1 + β2xi,t + β3ct + µi + υi,t (6)
i = 1, ..., 10; t = 2008i, ..., 2010ii
where yi,t and yi,t−1 correspond to each housing project markup in period t and t− 1
respectively. xi,t is a concentration market measure, which in this case is the sales shares
in t. ct is a business cycle variable, which is constant for all projects in a given period of
time. µi corresponds to a fixed effect for every project, which objective is to capture the
effects of particular features that do not change over time. υi,t is a stochastic shock.
The main aim is to test the business cycle variable impact on projects markups. Its
parameter value is β3. This parameter value will indicate if the empirical evidence is con-
sistent with either a countercyclical or a pro-cyclical behavior. Either a negative IMACEC
parameter value or a positive unemployment rate parameter value will be evidence of a
countercyclical behavior and the opposite of a pro-cyclical one.
The inclusion of the lagged dependent variable as a regressor enriches the specification.
The latter is due to he fact that this procedure allows to incorporate in an explicit way
the dynamic process that generates the project markups.
Machin and Van Reenen (1993) propose this methodology based upon a oligopoly
theoretical model developed by Cowling and Waterson (1976).
To test the lagged markups significance will allow to verify if the housing projects
markups show some inertia, as the Machin and Van Reenen (1993) model has predicted.
To obtain each project’s markups we have used the following definition:
yi,t =APi,tQi,t − Ci,t
Ci,t
(7)
where APi,t is the average price of project i in quarter t, Qi,t corresponds to project
i total sales in quarter t and Ci,t represents the project i total costs in quarter i. We
are aware about the fact that markups should be calculated using the marginal cost,
however because we have just aggregated cost information for each project (a description
of the data is given in the next subsection) we have used this formulation as a proxy.
10
Besides, as microeconomics theory indicates, long-run marginal and average costs are not
very different. Consequently, the usage of average costs for pricing may be regarded as a
reasonable approximation of marginal cost decision making.
2.4 Data
In hedonic price modeling there are three kind of information that is commonly used:
dwelling characteristics, location characteristics and environmental characteristics. Re-
garding the first one, the information used corresponds to the sales of new dwellings made
during 2008. In total there are 17,696 geo-referenced observations. For each one of them
there is information about price, in CPI-Indexed Unit of Account, UF (1 UF is about
US$20), type (either flat or house), surface in square meters, number of bathroom and
number of rooms. This information was bought from COLLECT GFK, a market research
company which elaborates every three months a Real Estate Market micro-database con-
taining the information mentioned above.
Regarding location characteristics, from Carabineros de Chile (Chilean Police) the
number of crimes committed in the dwelling neighborhood were obtained. Besides, using
the geo-referenced information, the distance from dwellings to the nearest green area,
urban highway access, urban highway (not to the access but the highway itself), cultural
center, school, police station, hospital and central business district, were calculated.
In relation with the environmental characteristics, from the CONAMA (the Chilean
environment agency), the records from the Santiago 11 measurement stations were ob-
tained.
For the tacit collusion test the data base used contains information about housing
projects since the first quarter of 2008 until the second quarter of 2010. Every record has
information about the total dwellings supplied, dwellings prices, size, number of bedrooms
and number of bathrooms. Also, it contains information about each projects sales by
quarter. This data base has been bought from COLLECT GFK. As it can gathered
the observations in this database correspond to housing projects, meanwhile that in the
database used for the hedonic model the observations are the housing units.
11
The construction costs were calculated using the construction cost per square meter
established by the Minister of Housing in order to charge the construction taxes.
The capital cost were calculated using the information about interest rates provided
by the Central Bank of Chile. From the same institution were obtained the business cycle
variables. We have used two different cycle variables. One is the unemployment rate, and
the second one is the activity index’s (IMACEC) deviation from its tendency.
3 Santiago Housing Market
The Chilean housing market has experienced an important boost during the last decade.
As a point of fact, in 2007 almost 50,000 dwellings were sold, which is the country highest
number of sales ever recorded in one year —in 2008 the financial crisis effects started
to be observed—. Figure 1 shows the seasonally-adjusted series of sales, where it can
be appreciated the significant growth and the posterior fall due to the sub-prime crisis,
which left the sales level slightly beneath the 2003 one.
There are several reasons behind this boom, such as the sustained country economic
growth, the real salary increase, 64.4% between 1980 and 2005, the subsidies given by
the State —almost two thirds of the housing production is made with the State sup-
port (Temino, 2007)—, and a deep and sophisticated mortgage market —the number of
mortgages grew 100% from 1998 to 2003—. As Uprah and Marcano (2008) have pointed
out, the key factor that have been essential for the mortgage market development are the
introduction of an inflation adjusted index unit (UF), the 1980 pension reform which cre-
ated a privately run compulsory capitalization system and an active process of innovation
in financial products that facilitated the transition from primary to secondary market
of mortgages. The latter has been characterized by three main instruments: Mortgage
Bills, Endorsable Mortgages and Non-Endorsable Mortgages. All these instruments are
indexed to the consumer price index, and have long maturities. These elements have had
as a consequence an expansion of the funds available and a fall of the interest rate.
12
Figure 1: Total sales (source: Chilean chamber of construction)
This high level of sales has been accompanied by a massive offer increase (see Figure 2),
which has been supplied by an important quantity of firms. These authors have identified
in 253 different developer in Santiago during 2010, and a Herfindhal index of 1.1%, which
indicates a low level of concentration. The latter has been used as an argument supporting
the idea that the Chilean housing market is very competitive.
As mentioned, the predictability of demand conditions affects the ability to sustain
collusion. Intuitively, tacit collusion will be easier to sustain in booming markets, when
future profits are expected to be high and, hence, the expected cost of retaliation is
also high. Conversely, collusion is more difficult to sustain in declining markets because
the short run gain of defecting will tend to compensate the limited expected cost of
future retaliation. As Rotemberg and Saloner (1986) and Haltiwanger and Harrington
(1991) state the strategic behavior outlined above has as an empirical implication that
tacit collusion will be less sustainable in markets that are subject to demand fluctuations
13
Figure 2: Dwellings supply (source: Chilean chamber of construction)
especially when they are predictable (see also Ivaldi et al. (2003) for a clear discussion
of this issue.) A second testable implication is that tacit collusion, under predictable
demand shocks and relatively homogeneous players will tend to generate a pro-cyclical
pattern of mark-ups. Therefore, we studied the sales dynamic pattern. Specifically, we
analyzed the demand shocks’ persistence and we found that they have an AR(1) behavior,
which implies the presence of a relatively predictable shocks. The results of this regression
are shown in Table 1. The dependent variable is the level of sales in quarter t, and after
proving different structures according to the information provided by the correlogram,
the model that fitted the best was the AR(1).
Albeit according to economic theory it is true that a great number of suppliers can
improve the market level of competition, it is also true that even in this case competition
cannot be guaranteed due to the potential emergence of tacit collusion. As a matter of fact
there are some facts related to this market behavior that would be interpreted as a lack of
14
Table 1: Estimation results : regress
Variable Coefficient (Std. Err.)
lag1 0.617∗∗ (0.150)
Intercept 881.193∗ (337.076)
N 32
R2 0.36
F (1,30) 16.844
competition. Particularly, during the sub-prime crisis, housing demand experienced and
important fall (see figure 1), nonetheless prices did not show the same behavior (see figure
3), as it could be expect when facing a highly inelastic supply. Instead, special sales during
limited periods of time were observed. Every two months special sales lasting just 72
hours were implemented by developers, which indicates a time-based price discrimination,
something that only would be done when firms have some extent of market power.
Figure 3: Hedonic Price Index (Authors’ own calculation)
15
However, an accurate housing market analysis must be based on sub-markets instead
of a whole city analysis. According to Alkay (2008) in a segmented housing market,
housing price structure is different in each segment and the whole city market does not
represent the housing services price. Consequently, the number of firms acting in every
sub-market will be much smaller, which implies a greater chance of observing low levels
of competition.
3.1 Hedonic Price Regressions and Santiago Housing sub-markets
The main goal of the present investigation, as it has been discussed above, is to test tacit
collusion in the Santiago Housing Market. However, as it was explained early on, in order
to perform this test in a more accurate way to identify sub-markets is needed, due to the
segmentation that this kind of markets present. Once this has been done, the test can be
performed for each one of these sub-markets. Consequently, before testing tacit collusion,
the sub-markets must be identified. In order to do that, the first step consist in running a
hedonic price regression, which allows to identified the main variables explaining dwelling
prices, by decomposing them into their characteristics implicit prices. Once the latter
has been done, the city blocks are pooled together following a socioeconomic criterium.
Consequently, all the blocks with the same average socioeconomic characteristics are clus-
tered together. Then, a hedonic regression is performed for each one of the proposed
sub-markets. Afterwards, a Chow test is performed to see if the parameters values that
have been obtained for every proposed sub-markets are different each other. If the test
results indicate that the parameters are different, then the the hypothesis that the Sub-
market are indeed different Sub-markets is not rejected. The next subsections explain
with more details the data, their sources and the process used to find the sub-markets.
3.1.1 Grouping process
The criterion used to cluster the blocks has been based upon the households socioeconomic
level. This has been obtained by the Adimark GFK using the 2002 census information.
16
This classification is based in two main variables: education level of the household head
and goods possession. The goods considered are: shower, TV, refrigerator, water heater,
microwave, cable TV, computer, internet and car. According to these two variables,
households were classified into five different groups called ABC1, C2, C3, D and E, ordered
in a decreasing way according to the socioeconomic level. For instance, if a households
has a head with no education and it has none of these goods, then it is classified as E,
but if a households has a head with more than 13 years of education and it possesses all
the goods mentioned above, then it is classified as ABC1. Following this procedure it has
been found that in Santiago 11.3% of the population corresponds to ABC1, 20.1% to C2,
25.6% to C3, 34% to D and 8.5% to E. The highest concentration of ABC1 households
is in the north east end of the city; the C2 and C3 groups are located in city centre, and
the D and E groups are located in the south and north ends. Figure 4 shows the spatial
distribution of these groups, being the ABC1 the blue ones, C2 the light green, C3 brown,
D orange and E red.
Every block with more that of 50% belonging to one particular group has been con-
sidered as a block of this group, i.e. if more than the 50% of households of one block
are ABC1, then this block is considered as ABC1. All the blocks belonging to the same
socioeconomic groups has been considered as one potential sub-market, even if they are
not contiguous. The socioeconomic groups D and E have been considered as 1 potential
sub-market, because the number of transactions related to these two groups is small, with
just 32 observations. Th reason is that although group D participation in the total popu-
lation is heigh, regarding housing markets the most of them participate in social housing
programs instead of private markets like the one that are studied here.
3.1.2 Hedonic regressions and sub-markets identification
Once the sectors have been identified, the next step consist in trying to cast light upon
if these sectors belong to the same housing sub-market or not. In order to do that first a
hedonic regression is performed to identify the variable explaining dwellings prices. Then,
a regression for each one of these sectors is run. Given the common hedonic models’
17
Figure 4: Spatial distribution of socioeconomic groups
problems of multicollinearity and heteroscedasticity, the regressions were made using the
heteroscedasticity and autocorrelation consistent covariance matrix.
Table 2 presents the variables used. Finally, a Chow test is performed to see if potential
sub-markets parameters estimated are statistically different each other. If they are is
because the potential sub-markets are effectively different sub-markets. Table 3 shows
the whole city hedonic regression results and table 4 shows the regressions results for each
sub-market.
After the tests for every potential sub-market have been performed, the results indicate
that submarkets 1, 2, 3 and 4 correspond to different sub-markets. the same sub-market,
i.e. that markets for ABC1, C2, C3, and D and E are different each other.
18
Table 2: Variables
Variable Description
valoruf house price in UF
tipo a dummy variable which takes the value 1 if the dwelling is a house
velkmhr the average speed, by car, from the house location to the city center
tiempomin the minimum time needed to reach, by car,
the city center from the house location
valorsuelom2 land price in UF
gsecodigo a code that indicates the potential sub-market where the house is located
metraje dwelling surface in square meters
bao number of bathroom
total delitos number of crimes committed in the dwelling municipality
d areas verdes distance in meters to de nearest green area
area verde2 square distance to the nearest green area
d acceso autopista distance in meters to the nearest urban highway entrance
d autopista distance in meters to the nearest urban highway
d centro comercial distance in meters to the nearest comercial center
d colegio distance in meters to the nearest school
col2 square distance in meters to the nearest school
d comisaria distance in meters to the nearest police station
com 2 square distance in meters to the nearest police station
d hospital distance in meters to the nearest hospital
hosp2 square distance in meters to the nearest hospital
d metro distance to the nearest underground station
vmh average maximum value of air pollution measure by the three
nearest pollution measurement stations
d subcentro distance in meters to the nearest central business district
19
Table 3: Estimation results : regress
Variable Coefficient (Robust Std. Err.) p-value
tipo -.1124945 .0064871 0.000
velkmhr -.0028969 .0001867 0.000
tiempomin .0009739 .0003822 0.010
gsecodigo -.2478184 .0032129 0.000
valorsuelom2 .0139352 .0003129 0.000
metraje .0052485 .0001597 0.000
bao .1112603 .0055585 0.000
total delitos -3.39e-06 1.69e-07 0.000
d areas verdes 1.83e-06 4.72e-06 0.684
d acceso aautopista -.0000231 9.62e-06 0.012
d autopista .0000107 9.25e-06 0.229
d centro comercial -8.36e-06 1.39e-06 0.000
d colegio .0000804 7.39e-06 0.000
d comisaria -4.27e-06 1.98e-06 0.029
d hospital 7.18e-06 1.35e-06 0.001
d metro 3.52e-07 2.74e-06 0.895
vmm -.0044038 .0003034 0.000
metro2 -2.53e-10 2.75e-10 0.347
cons 8.215717 .0272126 0.000
Number of obs 12202 R-squared 0.9462
Root MSE .14695
20
Table 4: Submarkets Estimation results
Variables S1 S2 S3 S4
tipo -.0695365* -.1785544* -.1654532* -.0320305
velkmhr -.0024668* -.0044234* -.0014514* -.0014515*
tiempomin .0041538* .0004031 -.0030328* -.0074586*
valorsuelom2 .0221933* .0151692* .0232607* .0101972*
metraje .0047248* .0077098* .0106327* .0065206*
bao .0934039* .0285029* .0459215* .0934484*
total delitos -.000026* -4.71e-06* -2.94e-06* 6.00e-08
d areas verdes -4.58e-06 .0000305* 9.34e-06 -.0000216
d acceso autopista -.0000347 -.0001504 * -.0001033* .0001358
d autopista .0000335 .0001193* .0000821* -.0001238*
d centro comercial 4.65e-06 3.95e-06 6.56e-06* -3.09e-06
d colegio .0000209 .0001056* -.0000577* .0001585*
d comisaria -.0000121* 1.67e-06 2.70e-06 4.55e-06
d hospital -.0000292* -8.33e-06* 4.59e-06* 4.21e-06
d metro 1.08e-06 .000014* 2.00e-06* -.0000522*
vmm -.0312665* -.0018121* -.0063237* .0006588
metro2 5.28e-10 -4.21e-09* -2.11e-09* 5.41e-09*
cons 10.26064* 7.576059* 7.357675* 6.654473*
Obs. 2884 2645 5854 819
R2 0.8815 0.7284 0.6635 0.3253
21
3.1.3 Sub Markets Characteristics
For characterizing sub-markets we focused the analysis on: concentration, markups,
projects size and firms multi market presence. The reasons for proceeding in this way
are, first, because any attempt for understanding a market structure demands a concen-
tration analysis to have an idea of how this market works. Second, as we want to test
the markups behavior, it is reasonable to try to understand its main characteristic before
performing any econometric analysis. Third, the projects size would give us an idea about
the potential presence of entry barriers, bigger average projects would be evidence of the
latter. Besides, from the degree of dispersion one can conclude about how homogenous
projects are: more homogenous projects will be an indication of similar cost structures,
which, in turn, would help the emergence of collusion (Ivaldi et al., 2003). Finally, the
multi-market competition facilitates the emergence of collusion (Bernheim and Whinston,
1990).
Concentration. For analyzing the level of concentration of each sub-market we cal-
culated the Herfindahl index and the Concentration Ratio for the 8 largest firms. Table
5 shows these indices. As it can be seen, these indices values indicates low levels of con-
centration, except for the sub-market S4 which exhibits medium levels of concentration.
Table 5: Herfindahl Index
Sub-market S1 S2 S3 S4
Herfindahl Index 287.805 394.447 239.71 1487.603
Concentration Ratio (CR8) 33.88% 30.55% 26.21% 55.56%
Markups. Figure 5 shows the average markups per quarter for each sub-market. It
is possible to see that markups vary significantly over time. The biggest variation can be
observed for sub-markets S2, S3 and S4. Besides, they show a similar pattern. Submarket
S1 presents a relatively modest variation. The biggest markups correspond to submarkets
S2, S3 and S4, and submarket S1 has significantly lower average markups.
22
Figure 5: Quarterly Average Markups per Submarket (Source: Authors’ own calculation)
Projects’ size. Figure 6 shows the projects’ average size for each sub-market. On
average, sub-markets S2, S3 and S4 have the biggest projects, meanwhile sub-market S1
presents an average project’s size which is 25% of the sub-market S3 one. Besides the
average size, it is interesting to pay attention to the extent of dispersion, because the
smaller the dispersion the similar the projects are. Table 6 presents the coefficients of
variation for each sub-market. All of them are bigger than 1, which is an indication
of a high level of dispersion. However, sub-markets S2 and S3 exhibit clearly a lower
dispersion, hence, is possible to say that these two submarkets have more homogenous
projects than those in sub-markets S1 and S4.
Table 6: Submarkets Coefficient of Variation
Sub-market S1 S2 S3 S4
Coefficient of Variation 1.38 1.31 1.24 1.49
Multi-market contact. Multi-market contact can sustain collusion, first, because
increases the firms interaction, and second because would allow softening asymmetries
that arise in individual markets (Ivaldi et al., 2003). We identified 244 different firms in
23
Figure 6: Average Projects Units Supply per Quarter
the whole market. There are 105, 57, 144 and 10 firms in sub-markets S1, S2, S3 and
S4 respectively. There are 194 firms just in one sub-market. In Table 7 there is a list
of the ten largest firms, which altogether control 25% of the whole market, and their
multi-market presence. They are ordered according to their market share. Except firms
7 and 8, all of them are in more than one market. Firms 1, 3 and 6 are in all markets.
Consequently, it can be argued that there are some extent of multi-market contact between
firms, particularly in the case of the ten largest ones.
Summarizing, the four markets present a low level of concentration. However, as it has
been discussed, this factor is not conclusive regarding collusion and strategic behavior.
Markups seem to be high, because, according to theory, under either perfect or monopolis-
tic competition (in the long run in the case of the latter) they must be nil. This situation is
particularly true for sub-markets S2, S3 and S4. The projects’ size give information about
24
Table 7: Multimarket Contact
Sub-market S1 S2 S3 S4
Firm 1 X X X X
Firm 2 X X X
Firm 3 X X X X
Firm 4 X X
Firm 5 X X
Firm 6 X X X X
Firm 7 X
Firm 8 X
Firm 9 X X
Firm 10 X X X
two scopes: the possible existence of entry barriers and the extent of homogeneity. The
bigger the projects the higher the possibility of entry barriers, given the costs associated
to land, capital, machinery and specialized laborers. Besides, the extent of homogeneity
of projects’ size can give an idea about how different is the costs structure amongst firms.
The level of dispersion is high in all sub-markets, but sub-markets S2 and S3 exhibit the
lowest level of dispersion. Finally, we found some extent of multi-market contact. With
all this information in hand we can hypothesize that if there would be any collusion it
should be observed, with a higher probability in sub-markets S2, S3 and S4, because they
have higher markups and concentration indices, and lower levels of dispersion. However,
at this stage is not possible to reach any kind of conclusion because the key elements are
the dynamic and strategic behavior, to conclude something about this issue. Next section
will focus, precisely, on these elements of analysis.
25
4 Testing tacit collusion
The simultaneous inclusion of the lagged dependent variable and the housing projects fixed
effect as a regressor in equation 6 generates an endogeneity problem. As a consequence,
the OLS estimations will not be consistent. To deal with this issue the methodologies
proposed by Arellano and Bond (1991), and Blundell and Bond (1998) (AB and BB
hereafter) and introduced to the economic analysis by Caselli et al. (1996), will be used.
These approaches are modified versions of the Generalized Moments Method (GMM),
with the particular feature that instrumental variables are lags of the same explanatory
variables. To have a clearer idea about this issue let us consider the following model
yi,t = αyi,t−1 + x′
i,tβ + εi,t
εi,t = µi + υi,t
E[µi] = E[υi,t] = E[µiυi,t] = 0
Within the GMM framework, developed by Hansen (1982), Arellano and Bond (1991)
propose to differentiate the model variables to eliminate the fixed effect, which is one of
the endogeneity sources. After applying this process the model will be:
∆yi,t = α∆yi,t−1 +∆x′
i,tβ +∆υi,t
however, despite that the fixed effect has been eliminated by this procedure, ∆yi,t−1 ≡
yt−1 − yt−2 is still an endogenous variable, because is correlated with ∆υi,t = υi,t − υi,t−1.
Consequently the inclusion of instrumental variables is needed. If υi,t does not present
serial autocorrelation, then yt−2 will be a valid instrument to ∆yt−1.
By extension, yt−3, ...yT−1 will be also valid instruments. If the variable is prede-
termined and it is not endogenous, then yt−1 is added to the available instruments set.
Blundell and Bond (1998), argue that when the lagged dependent variable is close to 1,
the instruments used in AB are weak, which reduces the estimators efficiency. To address
this problem, the BB methodology propose to expand the set of orthogonality condi-
26
tions. Particularly, this methodology proposes to differentiatte the potential instruments
in order to make them exogenous to the fixed effect.
Thus, if changes in yi,t−1, are orthogonal to the fixed effect µi, i.e. E[∆yi,tµi] = 0, for
all i and t, and there is not serial autocorrelation in υi,t, then ∆yi,t−1 is a valid instrument
for yt−1. As in the AB case, the BB methodology also includes as valid instrumental
variables ∆yt−1,∆yt−2, ...,∆yT−1.
The results of the tacit collusion test for the whole system are shown in tables 8 and
9. The former shows the results when the cycle variable used is the IMACEC’s deviation
from its tendency (cycle), meanwhile that the latter does when the variables used is the
unemployment rate. The dependent variable is the projects markups in UF. MarkupL1
corresponds to the one period lagged markups, and share corresponds to the project
quarter sales’ share. Finally, hhi is the each sub-market Herfindahl index. The markups
have been calculated as the difference between the total sales and costs. The costs where
obtained as follows: first, the project construction total cost is calculated using the square
meter cost proposed by the Minister of Housing in order to tax the construction projects
plus the land cost; second, each quarter cost is calculated as the percentage of the total
cost of the quarter sales (for instance, if in a quarter the 30% of one project is sold,
then quarter cost will be the 30% of the total cost), plus the alternative cost of money,
calculated as the investment multiplied by the quarter interest rate.
As it can be observed, the cycle variables are significant for the system as a whole,
which means that there is evidence to sustain that the complete Santiago housing market
exhibits a behavior consistent with tacit collusion. However, as it was argued early on,
in order to conduct a more accurate analysis it must be done at sub-market level. Table
10 shows the test results for each sub-market when the cycle variable is the IMACEC’s
deviation from its tendency, and table 11 shows the test results for each sub-market when
the cycle variable is the unemployment rate. The effect of industrial concentration is
estimated to be significantly positive, even after controlling for firm’s market share.
The cycle variables are highly significant (1%) for submarkets S2 and S3, but they are
not in sub-markets S1 and S4. Besides, sub-markets S1 and S4 do not show evidence of
27
Table 8: Dep = markup
Variable Coefficient
(Std. Err.)
MarkupL1 0.5252904∗∗
(0.1155291)
share 1.385475∗∗
(0.3207587)
cycle 0.2360538∗∗
(0.0579852)
hhi 0.0064933∗
(0.0029987)
∗5% significant ∗∗1% significant
Obs. 635
χ2(4) 2308.14
28
Table 9: Dep = markup
Variable Coefficient
(Std. Err.)
MarkupL1 0.5563229∗∗
(0.1186751)
share 1.408355∗∗
(0.3272125)
unemployment -15.76556∗∗
(4.092308)
hhi 0.4018439∗∗
(0.1022059)
∗5% significant ∗∗1% significant
Obs. 635
χ2(4) 2213.33
Table 10: Panel Data Submarkets Estimation results
Variables S1 S2 S3 S4
MarkupL1 0.1574* 0.8952* 0.39110* 0.9696054
share 1.3051** 1.5949* 1.24658** 1.221154
cycle -0.44745 0.2855707* 0.361243** 0.4420557
Intercept 8.024029 -37.1419* -41.45957** -58.75087
Obs. 195 140 246 37
χ2(3) 21.53 13.55 24.79 2.08
* 5% significant ** 1% significant
29
Table 11: Panel Data Submarkets Estimation results
Variables S1 S2 S3 S4
MarkupL1 1.574344 0.8952319* 0.4880787** 0.9696054
share 1.305165** 1.594913* 1.283789** 1.221154
unemployment 3.160634 -20.17173* -22.9835** -31.2259
Intercept -23.93883 167.1723* 196.5385** 257.8348
Obs. 195 140 246 37
χ2(3) 21.53 13.55 21.14 2.08
* 5% significant ** 1% significant
inertia. Consequently, the hypothesis of tacit collusion is rejected for sub-markets S1 and
S4. This result in the case of sub-market S4 would be explained by the fact that it has
just 32 observations. This hypothesis cannot be rejected for sub-markets S2 and S3. This
result can have important welfare implications because these two sub-markets represents
the 63% of the sample, which means that most part of the population face a colluding
housing market.
5 Final remarks
The aim of this investigation has been to test for tacit collusion in housing markets. In
order to achieve this objective we have used a rich sales micro data base from Santiago,
Chile. Using this database we have implemented a two steps methodology. First we
have split the Santiago’s housing market into submarkets. The sub-markets have been
defined based on housing’s prices and households’ socioeconomic characteristics. We have
found 4 different sub-markets which we have called S1, S2, S3 and S4. Submarkets S1
corresponds to the well-off households, S2 and S3 to middle income households and S4 to
the poor families (but not the poorest of society which have not been considered in this
study). Secondly, we have undertaken a tacit collusion test for each sub-market. e have
not found evidence of tacit for submarket S1 and S4. However our sample is too small in
30
the case of submarket S4 to be sure about this result. As S2 and S3 sub-markets show
evidence of tacit collusion, the biggest part of the population must face a housing market
with a low level of competition which implies important welfare costs. Amongst them
the most important would be the following: first, prices will tend to be higher than those
that guarantee economic efficiency. Second, the dwellings supply will be lower than the
one that could be observed in a competitive environment. Third, an important group of
the population will not be able to access to the market. Fourth, as developers will not
have incentives for differentiation, then households will have a lower chance of finding
an unit that fully satisfies their requirements. Fifth, developers will not have incentives
to improve dwellings quality. Sixth, given the higher prices, developers will extract rents
from households, which belong to the lower income segment, hence there will be a problem
of income distribution. Finally, developers will not have incentives to provide information
about dwellings quality and main characteristics, increasing searching costs and reducing
the households possibility of purchasing in an informed way.
Given the costs mentioned above, it is important for policymakers to try to design
policies focus upon the improvement of housing market competition. An important thing
to do is to generate an accurate micro database of housing projects and sales, identify-
ing dwellings features (size, location, amenities, etc.), transaction prices and households
characteristics. This information will allow to identify sub-markets, which must be the
geographical unit of analysis. Also, it is important to have detailed information about
the housing projects, like size, quality, costs, etc. Besides, all this information must be
geo-referentiated. Having a database with these characteristics will allow to analyze the
market behavior and to study the level of competition at a sub-market level, like it has
been done in this research.
Once this database has been collected, it must be processed and the results must be
freely available. The households free access to this sort of information will reduce their
searching costs, which is particularly important for lower income families. Consequently,
households will be able to compare quality and prices which, in turns, can encourage
competition. It is important to keep in mind that if policymakers want to facilitate this
31
comparison, the analysis and information processing must be done using the hedonic
methodology. Otherwise this comparison will not be valid.
It is important also to revise the subsidies scheme. In the Chilean case, the most
part of the subsides given have been focused upon those families belonging to S2 and S3
markets. As a matter of fact, according to Simian (2010), 49% of the total of subsidies
has been given to households belonging, in accordance with sub-markets classification
made here, to sub-markets S2 and S3. Consequently, albeit these subsidies aim has been
to facilitate the access of these families to the housing market, they have been maybe
funding extra-normal profits. Therefore, after doing a competition analysis, subsidies
would be given on the basis of the competition observed.
As the housing market is an important economy engine it is relevant to be cautious
in order to avoid the implementation of a policy which a side effect could be the lack of
interest of investors and developers in participating in the market. Due to the latter the
policy recommendations given here have been focus upon monitoring market behaviour,
the households access to the information and the improvement of the system of subsidies
instead of proposing regulations which can negatively affect not just the housing and real
estate market but the economy as a whole.
References
Alkay, E.: 2008, Housing submarkets in istanbul, International Real Estate Review
11(1), 113–127.
Arellano, M. and Bond, S.: 1991, Some tests of specification for panel data: monte-
carlo evidence and an aplication to employment equations, Review of Economic Studies
58, 277–297.
Bernheim, B. and Whinston, M.: 1990, Multimarket contact and collusive behavior,
RAND Journal of Economics 2(1).
32
Blundell, R. and Bond, S.: 1998, Initial conditions and moment restrictions in dynamic
panel data models, Journal of Econometrics 87, 115–143.
Caselli, F., Esquivel, G. and Lefort, F.: 1996, Reopening the convergence debate: A new
look at cross-country growth empirics, Journal of Economic Growth (3), 363–89.
Ching, S. and Fu, Y.: 2003, Contestability of the urban land market: an event study of
hong kong land auctions, Regional Science and Urban Economics 33(6).
Cowling, K. and Waterson, M.: 1976, Price cost margins and market structure, Economica
43, 267–274.
Fabra, N.: 2006, Collusion with capacity constraints over the business cycle, International
Journal of Industrial Organization 24.
Goodman, A. and Thibodeau, T.: 2003, Housing market segmentation and hedonic pre-
diction accuracy, Journal of Urban Economics .
Green, E. and Porter, R.: 1984, Noncooperative collusion under imperfect price informa-
tion, Econometrica (52), 87–100.
Haltiwanger, J. and Harrington, J.: 1991, The impact of cyclical demand movements on
collusive behavior,, RAND Journal of Economics 22(1), 89–106.
Ivaldi, M., B., J., Rey, P., Seabright, P. and Tirole, J.: 2003, The economics of tacit
collusion, Technical report, Final Report for DG Competition, European Commission.
Jones, C., Leishman, C. and Watkins, C.: 2004, Migration linkages between urban housing
submarkets: theory and evidence, Housing Studies 19, 269–283.
Machin, S. and Van Reenen, J.: 1993, Profit margins and the business cycle: evidence
from UK manufacturing firms, Journal of Industrial Economics 41(1), 29–50.
Maclennan, D.: 1977, Some thoughts on the nature and purpose of house price studies,
Urban Studies (14), 59–71.
33
Rapson, D.: 2009, Tacit collusion in the 1950ıs automobile industry? revisiting bresnahan
(1987), Working paper, Department of Economics, University of California, Davis.
Rosen, S.: 1974, Hedonic prices and implicit markets: product differentiation in pure
competition, Journal of Political Economy 82, 666–677.
Rotemberg, J. and Saloner, G.: 1986, A supergame-theoretic model of business cycles and
price war during booms, American Economic Review (76), 97–109.
Royuela, V. and Vargas, M.: 2009, Defining housing market areas using commuting and
migration algorithms: Catalonia (spain) as a case study, Urban Studies 46(11).
Schnare, A. and Struyk, R.: 1976, Segmentation in urban housing markets, Journal of
Urban Economics 3, 146–166.
Simian, J.: 2010, Logros y desafıos de la polıtica habitacional en Chile, Centro de Estudios
Publicos 117.
Straszheim, M.: 1975, An Econometric Analysis of the Urban Housing Market, NBER.
Taltavull de la Paz, P.: 2001, Economıa de la construccion, Civitas.
Taylor, L.: 2008, Theoretical foundations and empirical developments in hedonic modeling,
Vol. Hedonic methods in housing markets, Springer, chapter 1, pp. 15–37.
Temino, I.: 2007, El mercado hipotecario en Latinoamerica: una vision de negocio, an-
tecedentes y oportunidades de desarrollo, ESIC editorial.
Uprah, I. and Marcano, L.: 2008, Chile’s housing finance: A story of success?, Housing
Finance International .
34